The Pitfalls of D&D Beyond Data

FrogReaver said:

IMO That sounds like a great start to a brand new thread.

Click to expand...

FrogReaver said:

When you start posting in this thread with accusatory posts like the one above.

And flat out dismissive posts like the one directly below





And argumentative posts like the one above

And another dismissive posts like the post directly below



And more flat out dismissive posts like the 2 directly below





Anyways, after all THAT (and more), why do you think I care one bit about engaging you on something that is only tangentially related the topic of this thread (like what value can be had from the D&D Beyond dataset that none of us will ever actually be able to see)? Hint: I don't care about your tangent and at this point any goodwill I may have otherwise had toward discussing that particular tangent has been used up in my documented dealings with you above.

I mean seriously... it's almost like you've had it out for me specifically since you started posting in this thread. So I'll ask nicely, if this is a vendetta you have out for me, please drop it.

Click to expand...

FrogReaver said:

IMO That sounds like a great start to a brand new thread.

Click to expand...

TheSword said:

That’s a pretty dismissive response to the questions I posed.

Click to expand...

My post was an attempt to take your concerns and see how those pitfalls can be overcome to draw useful conclusions. Is that not relevant?

Click to expand...

The personal arguments seems to have taken over the discussion of what is an interesting topic.

Click to expand...

Bacon Bits said:

If the thread only allows negative criticisms of the data and doesn't allow any counter argument to that, then this thread is just, "let's bash DDB because I don't like the conclusions I would have to draw from the data they publish if the data are accurate." That isn't a discussion or a debate. It's a circle-jerk.

Click to expand...

Mistwell said:

Now that we've read responses from one of the people with actual access to the entire raw data, and explanations for some of what they did and how it is in fact accurate, I find this entire thread bizzare at this point. It made sense to begin with, but continuing now as if the initial objections have not been answered and answered well seems weird to me.

Click to expand...

FrogReaver said:

Can you link everyone to where that is?

Click to expand...

BadEye said:

Not so. Out of users that have unlocked the entire PHB, we still see the non-variant human selected more or less as frequently as variant.

I posit this could mean a smaller percentage of the playerbase is concerned with mechanical "optimization" than those of us that comment on internet threads are.

Click to expand...

BadEye said:

No, but it is possible to have a higher percentage in a separate data sample that removes characters that have not reached a level appropriate to choose a subclass, which is what has happened here.

This data is intended to provide a broad view into relative popularity between all other individual subclasses.

Click to expand...

BadEye said:

I'm not going to have time to deep dive an explanation on this, but you're not thinking of the data sets how they should be considered.

Taking an incredibly simple example, let's go with the entire population is 100 and it is composed of only Fighters and Clerics.

There are 70 fighters and 30 clerics. The 70 fighters choose (let's say) among 10 subclasses and (let's say) equally for the example. That means that for those 10 fighters subclasses, they each only compose 7% of the total percentage of individual subclasses.

For the 30 clerics, let's say that 20 of those are Life Domain, and the rest are Knowledge Domain (10). This gives us a final breakdown of:

20% Life Domain, 10% Knowledge Domain, and 10 other subclasses at 7%. Even though there are only 30 total clerics compared to the 70 total fighters, Life Domain still comprises the highest individual percentage.

Of course mileage is going to vary on any of this. The intent is to demonstrate which subclass choices are popular comparatively. If someone chooses cleric, there's a pretty high percentage they're choosing Life, same as with Draconic sorcerer. Fighter is the most popular class overall, but they have a higher number of better-represented subclasses, therefore bringing their individual percentages lower.

It might be more helpful to consider the rankings by class I shared back in April. (This is not updated for current, but we will provide that in the near future. I couldn't figure out how to inline this, so I tried to attach.)

Click to expand...

BadEye said:

Yes...it is?

I guess this will be my last attempt to explain, this time with a question - do you see any single other cleric domain in the list on that chart?

No being the answer, the reason is such a high percentage of clerics choose Life Domain that the total of that subset is higher than all the other individual subclasses for all the other classes. I can confirm that over 60% of clerics on DDB are Life Domain, and clerics have the most subclasses of any class. Only sorcerers exhibit more single-subclass dominance with Draconic comprising 65% of all sorcerers (sorcerers just have a good bit fewer total number of characters).

Perhaps one other part you're missing is that the population for the subclasses chart already removes all characters without subclasses. Since it is looking only at relative subclass distribution, it only includes characters with subclasses. In other words, the subclasses chart does not use the same population as the classes chart.

I tried here. I can assure you that we have an actual statistician that is doing this stuff (not me) and that it's what we see. You are making some pretty dogmatic assertions about datasets you haven't seen in detail.

At the end of the day, you're right with some of your other comments - this information does not demonstrate any class or subclass is "better," more balanced, or anything else than another, no more than it used to mean when they shared World of Warcraft class or race distributions. Sometimes inferences can be made, but nothing is completely solid since we can't guarantee a character is being played versus just created (even though we attempt to do what we can there).

I hope you have a great rest of the weekend!

Click to expand...

BadEye said:

Again, very assertive for someone that does not have the complete picture of the data.

For instance, I have not mentioned anything about multiclassing, which impacts this greatly. Taking the barbarian class for a couple of levels is actually very popular for multiclass characters, and those would not end up having a barbarian subclass yet. When looking at subclasses only, those barbarians would be removed from the dataset.

We could absolutely remove any characters that are multi-class from these results, but there are pros and cons for doing so and we decided to keep them in for now.

So, having the actual data in hand, I will assert that it is correct, but as with any analysis uses some assumptions and parameters that you do not have access to for reverse engineering.

The goal of sharing these numbers is not to concretely establish camps out there in the community - it is an interesting exercise that could demonstrate player choices and trends.

I do appreciate your thirst for accuracy and ensuring that no one out there is being bamboozled!

Click to expand...

BadEye said:

The chart adds up to 100% (of course all circle charts do) - there just isn't labeling for the remaining percentages because there was no room for them to exist and still be legible.

Click to expand...

BadEye said:

This data is presented as a high-level look at distribution of race, class, and subclass selection for active characters on D&D Beyond. It serves that purpose just fine.

It could certainly "confuse" or "amuse" those who want to read too much into it or think it is trying to achieve a different purpose than it is.

Actual decisions could be (and are) made from this data. For instance, when looking at future subclass design, maybe the bar would be set at Life Domain and not at the least selected domain. It's a safe bet to say that draconic sorcerers "do well" in the community and other design should target that, or that perhaps people would want to know more about half-elf culture since they are so often chosen. The data never pretends to dig any deeper than that.

Click to expand...

BadEye said:

This is my point that no one here (except me) knows the actual dataset and many assumptions are being made. Multiclassing and homebrew subclasses are going to throw off any napkin math you all can do.

If we entirely removed multiclass characters, for instance, does this actually still give us the most accurate look at class popularity? Is a class still popular if it is chosen, even if for only a level or two?

We can absolutely remove multiclass characters (and we have before), but I can tell you the distribution doesn't actually change that much.

Click to expand...

BadEye said:

No worries - I appreciate passion.

Click to expand...

Mistwell said:

You may recall who it is by the way, since you insulted the guy, then apologized to him, resulting in that last response. Remember now?

Click to expand...

Mistwell said:

Now that we've read responses from one of the people with actual access to the entire raw data, and explanations for some of what they did and how it is in fact accurate, I find this entire thread bizzare at this point. It made sense to begin with, but continuing now as if the initial objections have not been answered and answered well seems weird to me.

Click to expand...